graceful-degradation

SKILL.md

Graceful Degradation with Helpful Messages

When optional services are unavailable, degrade gracefully with actionable fallback messages.

Pattern

Check availability at the start, cache the result, and provide helpful messages that explain what's missing and how to fix it.

DO

Check service availability early (before wasting compute)

Cache health check results for the session (e.g., 60s TTL)

Provide actionable fallback messages:

What service is missing

What features are degraded

How to enable the service

Continue with reduced functionality when possible

DON'T

Silently fail or return empty results

Check availability on every call (cache it)

Assume the user knows how to start missing services

Example: LMStudio Check Pattern

let lmstudioAvailable: boolean | null = null;
let lastCheck = 0;
const CACHE_TTL = 60000; // 60 seconds

async function checkLMStudio(): Promise<boolean> {
  const now = Date.now();
  if (lmstudioAvailable !== null && now - lastCheck < CACHE_TTL) {
    return lmstudioAvailable;
  }

  try {
    const response = await fetch('http://localhost:1234/v1/models', {
      signal: AbortSignal.timeout(2000)
    });
    lmstudioAvailable = response.ok;
  } catch {
    lmstudioAvailable = false;
  }
  lastCheck = now;
  return lmstudioAvailable;
}

// Usage
if (!await checkLMStudio()) {
  return {
    result: 'continue',
    message: `LMStudio not available at localhost:1234.

To enable Godel-Prover tactic suggestions:
1. Install LMStudio from https://lmstudio.ai/
2. Load "Goedel-Prover-V2-8B" model
3. Start the local server on port 1234

Continuing without AI-assisted tactics...`
  };
}

Fallback Message Template

[Service] not available at [endpoint].

To enable [feature]:
1. [Step to install/start]
2. [Configuration step if needed]
3. [Verification step]

Continuing without [degraded feature]...

Source Sessions

This session: LMStudio availability check with 60s caching and helpful fallback

174e0ff3: Environment variable debugging - print computed paths for troubleshooting